Data-driven artificial neural network formulated multi-factored expression for predicting construction duration in government projects

Purpose Project delays are a considerable challenge in the construction industry, often leading to financial disputes and complications. Traditional methods for estimating construction project duration are error-prone and frequently underpredict the total period due to hastily formulated activity timelines. This study aims to develop a reliable data-driven artificial neural network (ANN) approach and an accurate predictive model for estimating construction durations in government projects. Design/methodology/approach This study analyzes 111 government-approved projects to identify key variables such as project locality, project type, project cost, inclement weather, internal issues, external obstructions, unforeseen events and relativity. These variables are statistically analyzed and then trained, tested and validated using an ANN technique. The ANN model is developed using MATLAB, with a training:test:validation ratio of 70:15:15, from which the first-ever practically convenient explicit equation has been formulated to help construction practitioners make a quick and reliable prediction of project duration. ANOVA and Shapley impact analyses have been performed to showcase the significance of the proposed factors including their ranking of importance. Findings The study identifies key factors impacting project timelines and utilizes the ANN model to create a multi-variable straightforward yet precise mathematical formula for duration estimation. The model demonstrates a high degree of agreement between predicted and actual project timelines, with correlation coefficients of 0.99396 for training, 0.95618 for validation, 0.98057 for testing and 0.98225 overall. The prediction coefficient of determination is 0.929 and the mean absolute percentage error is 15.66%, indicating the model’s accuracy. All factors or their interactions display high significance ( p < 0.05) towards project duration with relativity, project cost and locality ranked as the top three influencing components. Originality/value Differing from standard black-box neural network models, our ANN approach produces a practically convenient single-line equation that can be applied with a basic hand calculator, making it accessible to engineers and project managers. This research highlights the critical importance of evidence-based methods in enhancing construction project management, ensuring that project estimates are dependable and easily applicable in practical situations.